RM新时代有限公司

Recently, CloudWalk Technology officially won the bid for the document image information extraction model construction technical service project based on multimodal large model of Shandong Branch of Postal Savings Bank of China (hereinafter referred to as "PSBC Shandong Branch"), and will jointly create a financial industry large model - Document Multimodal Large Model.

This is another benchmark case for CloudWalk Technology in the field of multimodal large models, and it is also a full affirmation of CloudWalk Technology's strength by PSBC Shandong Branch.

The focus of this cooperation is to solve the document image processing challenges in financial scenarios.

Financial scenarios involve many types of document images, including cards, bills, contracts, forms, reports, etc. Different types of documents also have a large number of subtypes, with problems such as complex and diverse layouts, large differences in content expression, large span of contextual information, and difficulty in collecting real data.

Traditional OCR and document parsing often require customized development for a specific subtype, resulting in high development costs and lack of generalization ability, while also having low semantic understanding of documents.making it difficult to cope with these challenges..

To overcome these difficulties, CloudWalk Technology and PSBC Shandong Branch joined hands to deeplyexploreindustry pain points. They optimized and upgraded the defects of traditional OCR and document parsing, and developed a document multimodal large model based on the CongRong Large Model.

The emergence of the document multimodal large model will greatly improve the generalization performance across different layouts, enabling it to adapt to various complex document layouts, and effectively handle bills, cards, contracts, forms, reports, etc. without fixed layouts.

It not only reduces customized development costs but also improves the applicability of the large model in different scenarios.

In addition, through the correlation between different modalities such as language and vision, the document multimodal large model has excellent positioning ability and multi-answer text extraction ability, enabling accurate extraction of non-continuous answers.

In terms of data, the document multimodal large model uses a self-developed document synthesis tool to achieve data simulation generation. For difficult scenarios, data synthesis and processing tools can also be used for adaptive training, thereby quickly and flexibly optimizing model effects, solving pain points in business data confidentiality and cost.

As a company committed to leading the new wave of artificial intelligence, CloudWalk Technology, with its world-leading and independently controllable core large model technologies, has repeatedly set world records in multiple fields such as language and vision. In the future, CloudWalk Technology will continue to uphold the spirit of innovation, providing better technologies and solutions for the financial industry and promoting the intelligent process of the industry.